abliteratedmodels.org

Qwen3.5-4B-Claude-4.6-Opus-abliterated

Qwen3.5-4B-Claude-4.6-Opus-abliterated is a refusal-ablated variant of Jackrong/Qwen3.5-4B-Claude-4.6-Opus-Reasoning-Distilled, published on Hugging Face by huihui-ai. It is 4B parameters, single consumer GPU class and apache-2.0 licence. It is distributed across 10 repositories in MLX, Transformers, GGUF and GGUF (imatrix) formats, totalling 5.8K downloads.

Abliteration (directional ablation)4BQwen4B – 10B5,83181

Specification

Model name
Qwen3.5-4B-Claude-4.6-Opus-abliterated
Publisher
huihui-ai
Parameters
4B
Hardware class
4B – 10B — single consumer GPU
Licence
apache-2.0
Task
Image text to text
Context window
Not documented
Ablation scope
Not stated by the publisher
Formats available
MLX, Transformers, GGUF, GGUF (imatrix), NVFP4
First indexed
15 Mar 2026
Last updated
13 Apr 2026

Ablation technique

How Qwen3.5-4B-Claude-4.6-Opus-abliterated was modified, and what that implies.

Abliteration (directional ablation)

The single residual-stream direction that mediates refusal is identified from harmful/harmless prompt pairs and projected out of the model weights.

Reported scope: Not stated by the publisher. Publishers frequently omit this, so verify behaviour empirically rather than assuming full coverage.

From the publisher’s model card

This is an uncensored version of Jackrong/Qwen3.5-4B-Claude-4.6-Opus-Reasoning-Distilled created with abliteration (see remove-refusals-with-transformers to know more about it). This is a crude, proof-of-concept implementation to remove refusals from an LLM model without using TransformerLens.

Running it

Commands are templates — confirm the exact repository and quant file before use.

Ollama (publisher-documented)
ollama run huihui_ai/qwen3.5-abliterated:4b-Claude
llama.cpp / GGUF
hf download mradermacher/Huihui-Qwen3.5-4B-Claude-4.6-Opus-abliterated-GGUF --local-dir ./qwen3.5-4b-claude-4.6-opus-abliterated
llama-cli -m ./qwen3.5-4b-claude-4.6-opus-abliterated/<file>.gguf -p "..."
Transformers
from transformers import AutoModelForCausalLM, AutoTokenizer

repo = "huihui-ai/Huihui-Qwen3.5-4B-Claude-4.6-Opus-abliterated"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(
    repo, torch_dtype="auto", device_map="auto"
)
vLLM
vllm serve huihui-ai/Huihui-Qwen3.5-4B-Claude-4.6-Opus-abliterated --trust-remote-code

Downloads and variants (9)

Every published repository of Qwen3.5-4B-Claude-4.6-Opus-abliterated, including quantised re-releases by other authors.

Security-team assessment

Derived from this model’s metadata and the publisher’s claims — not from benchmarks run by this site.

  • Qwen3.5-4B-Claude-4.6-Opus-abliterated will attempt offensive-security prompts that a hosted commercial model declines, which is what makes it usable for red-team corpus generation and for measuring what an unaligned model of this class produces.
  • Capability is inherited from Jackrong/Qwen3.5-4B-Claude-4.6-Opus-Reasoning-Distilled, not added by ablation — at 4B parameters, expect a small model’s reasoning and a high error rate on exploit detail.
  • The publisher does not state which layers were ablated, so assume nothing about refusal consistency — probe it directly.
  • Declared licence is apache-2.0, but the terms that bind you are Jackrong/Qwen3.5-4B-Claude-4.6-Opus-Reasoning-Distilled’s — a re-release cannot grant more than its parent.

Handling: run it isolated, put moderation in front of it if anyone outside your team can reach it, and verify the weights before loading — these are third-party artifacts. Full handling and licence guidance.

Metadata for Qwen3.5-4B-Claude-4.6-Opus-abliterated is collected from the public Hugging Face API and the publisher’s model card. This site does not host weights, is not affiliated with huihui-ai, and does not independently verify publisher claims. See responsible use.